Google Says Threat Actors Used AI Framework to Find and Exploit Vulnerabilities Autonomously

Summary
Google's Threat Intelligence Group says it found exploit_agent, a multi-agent framework designed to find vulnerabilities, develop exploits and execute them autonomously. Its findings also describe broader malicious use of AI tools.
Key points
- GTIG says it identified threat actors using an AI framework to automate vulnerability discovery and exploitation without human intervention.
- The framework, exploit_agent, was found on an exposed staging server and was designed to compare patched and unpatched web application code, identify flaws, develop exploits and execute them.
- GTIG says the server also held data stolen from victims and scripts for post-compromise activity, including reconnaissance and webshell deployment.
- GTIG observed Cairn used in activity targeting the US aerospace sector and PentAGI used for reconnaissance and vulnerability scanning across organizations in Asia.
- The report describes malicious use of LLMs for phishing, influence operations and LLMJacking, in which actors seek credentials to access and monetize AI services.
- GTIG expects malicious use of autonomous tools to grow, while noting that LLM use can introduce forensic noise and reduce operational reliability.
Article Details
- Event Type
- Google Threat Intelligence Group reported that threat actors are using AI frameworks to automate vulnerability discovery and exploitation, alongside broader malicious uses of generative AI.
- Impact
- The identified exploit_agent framework was designed to compare patched and unpatched web application code, find vulnerabilities, develop exploits, and execute them automatically. Its exposed staging server also contained stolen victim data and post-compromise scripts, including reconnaissance and webshell deployment. GTIG reported interest in French media and entertainment websites and the education sector. The article also describes AI-assisted activity targeting aerospace and organizations in Asia, phishing operations, influence campaigns, and LLMJacking that can drive substantial usage and billing increases.
People
Threat Actors
CALANQUE IONIran-linked APT reported to have adopted LLMs; the article identifies Calanque Ion as another name for APT42.LEAKY CASTLEIdentified as a Chinese APT that has adopted LLMs.MIDNIGHT NEPTUNENorth Korean hackers identified in connection with IT worker activity and attacks against cryptocurrency organizations; the article also gives UNC1069 as another name.UNC1069Name the article gives for Midnight Neptune, which it links to North Korean IT worker activity and attacks against cryptocurrency organizations.
Tools
Agent Zeroservices, such as ZeroHollow, an inference service used for exploit development and LLM routing, and Agent Zero, an open source agent framework which features autonomous code execution and dynamic tool generation,”ARTEXAttackers are using frameworks such as Artex, Cairn, CyberStrikeAI, and PentAGI to increase the velocity of their multi-stage intrusions, and rely on LLMs and agentic frameworks to reverse engineer patches at machineCairnAttackers are using frameworks such as Artex, Cairn, CyberStrikeAI, and PentAGI to increase the velocity of their multi-stage intrusions, and rely on LLMs and agentic frameworks to reverse engineer patches at machineCyberStrikeAIAttackers are using frameworks such as Artex, Cairn, CyberStrikeAI, and PentAGI to increase the velocity of their multi-stage intrusions, and rely on LLMs and agentic frameworks to reverse engineer patches at machineexploit_agentThe multi-agent framework, called exploit_agent, was discovered on an exposed staging server.HexStrike AIusing and experimenting with open source frameworks such as Cairn, PentAGI, CyberStrikeAI, and HexStrike AI.JieyouThe Jieyou phishing service, active since March 2026, has a model-agnostic chatbot that allows clients to configure phishing panels and modify templates.PentAGIAttackers are using frameworks such as Artex, Cairn, CyberStrikeAI, and PentAGI to increase the velocity of their multi-stage intrusions, and rely on LLMs and agentic frameworks to reverse engineer patches at machineZeroHollow“These observations additionally extend to the use of other tools and services, such as ZeroHollow, an inference service used for exploit development and LLM routing, and Agent Zero, an open source agent framework which
Countries
ChinaRelated: Trump Administration Moves to Block China From Indirect Access to American AI TechnologyFranceAccording to GTIG, evidence pointed to an interest in French media and entertainment websites, as well as the education sector.IranState-sponsored groups known to have adopted LLMs include multiple Chinese APTs, such as Leaky Castle; Iran-linked APT42, also known as Calanque Ion; and North Korean hackers supporting IT worker activity and runningNorth Koreamultiple Chinese APTs, such as Leaky Castle; Iran-linked APT42, also known as Calanque Ion; and North Korean hackers supporting IT worker activity and running attacks against cryptocurrency organizations, suchUnited StatesGTIG observed Cairn being used in activity targeting the aerospace sector in the US, and PentAGI being used for systematic reconnaissance and vulnerability scanning across organizations in Asia.
Industries
AerospaceGTIG observed Cairn being used in activity targeting the aerospace sector in the US, and PentAGI being used for systematic reconnaissance and vulnerability scanning across organizations in Asia.Critical infrastructure“For US critical infrastructure, the danger isn’t just more attacks, but attacks at machine speed.Cryptocurrencyas Calanque Ion; and North Korean hackers supporting IT worker activity and running attacks against cryptocurrency organizations, such as Midnight Neptune (UNC1069).EducationAccording to GTIG, evidence pointed to an interest in French media and entertainment websites, as well as the education sector.Media and entertainmentAccording to GTIG, evidence pointed to an interest in French media and entertainment websites, as well as the education sector.
